Skip to main content
A decision starts with a goal, the facts so far and at least two options. Your AI then answers one question at a time until none remain, records the choice with its reason, next action and how you will know it worked, and later reports how it went. On the free plan a decision is not kept after it is made; a paid plan keeps every decision, note and result. You do not call these tools yourself. You ask your AI in plain words, and it picks the tool. This page shows what each one does, so you know what to ask for.

List decision methods

list_methods · Reads only Lists the decision methods start_decision can use: the universal one (id “universal”, for any decision) and methods for particular subjects such as cooking, relationships, changing software and tennis. Returns each method’s id, name and what it is for; pass the id as start_decision’s method, or leave method out for the universal one. Needs nothing and changes nothing. It takes no inputs.
Example

Start a decision

start_decision · Changes things Opens a decision and returns its case_id and first question. Needs source_id (any unique text for this decision; the same source_id with the same content returns the same decision), goal (what is being decided, in the person’s words), evidence (the facts so far, each {ref, claim}, where ref is a short name the answers cite), and at least two options ({id, label}). method (optional) is an id from list_methods. Then answer_question until no question remains, and decide. It decides nothing by itself. Inputs
Example

Get the next question

next_question · Reads only Reads a decision’s current question by case_id: the question with its answer_schema, which checks were selected or left out, and whether it is ready to decide (then call decide). Use it to pick up a decision where it was left. It changes nothing. Inputs
Example

Answer a question

answer_question · Changes things Saves the answer to the current question. The answer follows the answer_schema in the question: each listed field, evidence_refs naming the case evidence, and an optional scratchpad note. For fields listed in ask_others, answered_by records where the answer came from: “person”, or the memory, file or conversation you found it in. When the question lists a checklist, checklist gives, for each check id, how the answer handles that check or why it does not bear on this decision. usage (optional) reports what answering this step cost you: model, model_calls, input_tokens, output_tokens, or anything else countable. Inputs
Example

Revise the evidence

revise_evidence · Changes things Changes the evidence of an open decision: add new facts ({ref, claim}), correct a claim that was wrong ({ref, claim}), or retract a fact that turned out false ({ref, because} or the ref). A retracted fact can no longer be cited; every change is logged. On a free plan use the case_id from the latest reply. Inputs
Example

Add a note

add_note · Changes things Adds a note to one decision (case_id), optionally tied to one question (step_id): a thought, a fact found along the way, anything worth keeping with that decision. Returns the note and how many notes the decision now has. It is only for decisions: for what the person is pursuing use record_pursuits, for what was said use record_messages, for what was learned about a file use record_file_results. Inputs
Example

Record the decision

decide · Changes things Records the choice once every question is answered: option_id, reason, evidence_refs, next_action, success_measure, move when the move chosen is narrower than the option (the export then shows the move), and every field the last reply lists in decision_fields, each by its name. Sending the same decision again changes nothing and says already_decided. Returns the next action and what the whole run cost. usage (optional) reports what deciding cost you, the same way as for answers. Inputs
Example

Report how it went

report_result · Changes things Records what actually happened after a decision (case_id) or a tool run (task_id), as the person tells it: result (met, failed or unknown), summary in their words, and answered_by (“person” when the person said it, or who saw it). Ask the person before reporting: a tool’s contract passing only means its output had the right shape, and tools are ranked on these reports. Stored as the person’s report, apart from checked results. Returns the decision or task with its reports. Inputs
Example

List my decisions

list_decisions · Reads only Lists this account’s saved decisions, newest first: each with its id (the case_id other decision tools take), goal, status and the chosen option. Use it to find a decision to continue, export or report on. limit caps how many (default 20). It changes nothing. Inputs
Example

Export a decision

export_decision · Reads only Returns one whole decision (case_id) as a readable copy, as format “markdown” (the default) or “json”. It changes nothing and does not send it anywhere. Inputs
Example

Find similar decisions

related_decisions · Reads only Finds decisions like this one (case_id): the person’s own earlier ones (yours) and, where available, how other people decided in similar situations (others, anonymized, never naming anyone). Each shared decision offers questions to consider, linked to the anonymous source through read_shared_decision. Reading earlier decisions is part of the paid plan; on another plan it says so and returns none. It changes nothing. Inputs
Example

Read the shared decision behind a question

read_shared_decision · Reads only Reads the anonymous shared decision behind a suggested question. Use the shared_decision_id in the question’s source. If its owner stops sharing it, the source is no longer available. Inputs
Example

What has helped

what_helped · Reads only What has helped this person decide well, from their own saved decisions: how many hit their goal (with how many have a result yet), time from deciding to a checked result, whether their latest decisions are faster, cheaper or more often right than earlier ones, and which of the method’s questions most often changed their answer and how those decisions turned out. Every share comes with its counts; below the minimum it says there is not enough data yet. It takes no inputs.
Example

Score known answers

score_candidates · Reads only Picks from answers you already have, instead of generating one: give the situation (state), optionally the question (instructions), and criteria as { candidate id: description }, 2 to 200 candidates. Returns each candidate’s probability, the choice, a confidence from 0 to 1, and escalate: true when the confidence is below the line, meaning a slower check should decide. Fast and cheap; best for routing and shortlisting, not for judging whether something really happened. Inputs
Example
Last modified on October 1, 2026